Search

International audience; We advance a method of thesaurus enrichment, based on the extraction of coordinations in a domain-related corpus. Our hypothesis is that there is a semantic homogeneity between the conjuncts located in a coordination. We conducted an experiment that allowed us to evaluate the effectiveness of our method. This experiment aims to enrich the concept hierarchy of a French agricultural thesaurus named French Crop Usage (FCU), thanks to the texts of the Plant Health Bulletins (PHB). The FCU thesaurus is published on the Web using the SKOS model.

International audience; Predictive modeling, often called “predictive analytics” in a commercial context, encompasses a variety of statistical techniques that analyze historical and present facts to make predictions about unknown events. Often the unknown events are in the future, but prediction can be applied to any type of unknown whether it be in the past or future. In our case, we present some experiments applying predictive modeling to the usage of technical terms within the NLP domain.

Integrated approach to identify terms for monitoring disease emergence on the web.Terms are extracted automatically from disease outbreak web pages.Domain experts identify the terms relevant to characterise a disease emergence.Relevant terms are used as queries to mine the web. Timeliness and precision for detection of infectious animal disease outbreaks from the information published on the web is crucial for prevention against their spread. The work in this paper is part of the methodology for monitoring the web that we currently develop for the French epidemic intelligence team in animal health. We focus on the new and exotic infectious animal diseases that occur worldwide and that are of potential threat to the animal health in France.In order to detect relevant information on the web, we present an innovative approach that retrieves documents using queries based on terms automatically extracted from a corpus of relevant documents and validated with a consensus of domain experts (Delphi method). As a decision support tool to domain experts we introduce a new measure for ranking of extracted terms in order to highlight the more relevant terms. To categorise documents retrieved from the web we use Naive Bayes (NB) and Support Vector Machine (SVM) classifiers.We evaluated our approach on documents on African swine fever (ASF) outbreaks for the period from 2011 to 2014, retrieved from the Google search engine and the PubMed database. From 2400 terms extracted from two corpora of relevant ASF documents, 135 terms were relevant to characterise ASF emergence. The domain experts identified as highly specific to characterise ASF emergence the terms which describe mortality, fever and haemorrhagic clinical signs in Suidae.The new ranking measure correctly ranked the ASF relevant terms until position 161 and fairly until position 227, with areas under ROC curves (AUCs) of 0.802 and 0.709 respectively.Both classifiers were accurate to classify a set of 545 ASF documents (NB of 0.747 and SVM of 0.725) into appropriate categories of relevant (disease outbreak) and irrelevant (economic and general) documents.Our results show that relevant documents can serve as a source of terms to detect infectious animal disease emergence on the web.Our method is generic and can be used both in animal and public health domain.

International audience; The research presented in this paper explores the possibility of enriching terminological databases through the analysis of recent scientific publications. Our main concern is to evaluate how useful automatic term extraction can be to a human expert. To carry out our experiment, we constructed two corpora of recent scientific papers in two different sub-domains of the bio-medical sciences. Then we proceeded with three steps: automatic term extraction and ranking from a set of corpora of scientific papers; evaluation of the overlap of the candidate terms (CTs) extracted from the corpora and those present in the multidisciplinary terminology portal TermSciences; and evaluation by domain experts of the three sets of the top 200 CTs extracted from the different corpora. To extract terms we used the Sensunique Platform, a web based platform for building terminological resources. Our results show that only about 10% of the extracted CTs are present in the TermSciences resource, which means that many of the extracted CTs, if validated, could potentially be used to enrich the terminological database. Furthermore, the expert evaluation of the top 200 terms for each sub-corpus shows clearly that about 75% of these CTs are correct terms in the respective domains. This validates our ranking algorithm.

International audience; The increasing availability of parallel bilingual corpora and of automatic methods and tools for their processing makes it possible to build linguistic and terminological resources for low-resourced languages. We propose to exploit various corpora available in several languages in order to build bilingual and trilingual terminologies. Typically, terminology information extracted in French and English is associated with the corresponding units in the Ukrainian corpus thanks to the multilingual transfer. According to the used approaches, precision of the term extraction varies between 0.454 and 0.966, while the quality of the interlingual relations varies between 0.309 and 0.965. The resource built contains 4,588 medical terms in Ukrainian and their 34,267 relations with French and English terms.

Nous proposons une nouvelle méthode pour l’extraction de termes multi-mots à partir de publications scientifiques. Notre stratégie est fondée sur la combinaison de deux approches : une première liste de termes « candidats » est d’abord extraite à partir de critères de fréquence et de spécificité. Cette liste est ensuite classée suivant la position du terme dans le Résumé : (en ayant recours à un étiquetage de la valeur « argumentative » des phrases, selon une analyse de type text zoning). Cette approche permet de classer les termes en différentes catégories, et notamment de distinguer le vocabulaire conceptuel des éléments d’ordre méthodologique. Nous avons appliqué cette méthode à l’extraction des termes utilisés en traitement automatique des langues à partir de l’analyse d’un corpus (ACL Anthology) s’étendant de 1980 à 2008. Nous montrons ainsi qu’il est possible de suivre les méthodes utilisées, comment elles sont introduites dans le domaine, par quel type d’auteur et pour quel usage, etc. Nous observons ainsi plusieurs faits marquants de l’évolution du domaine sur une période de plus de 30 ans., We propose a new method to extract multiword expressions from scientific papers. Our approach is made of two major steps: a first list of candidates is extracted based on a score using frequency and specificity information. This list is then filtered based on the status of the term in the abstract of the scientific papers under investigation. These abstracts are annotated using a text zoning analyser. The terms are then classified in different categories according to the text zoning analysis: we make a difference between terms appearing in the method section of the abstract vs terms appearing in other zones. This method is applied to the ACL Anthology collection, containing the papers published by the ACL between 1980 and 2008. We show that the technique we use allows us to model interesting facts concerning the evolution of the domain and of the methods used in computational linguistics.

International audience; Semantic annotation using biomedical ontologies is required to enable data integration, interoperability, indexing and mining of biomedical data. When used to support semantic indexing the scoring and ranking of annotations become as important as provenance and metadata on the annotations themselves. In the biomedical domain, one broadly used service for annotations is the NCBO Annotator Web service, offered within the BioPortal platform and giving access to more than 350+ ontologies or terminologies. This paper presents a new scoring method for the NCBO Annotator allowing to rank the annotation results and enabling to use such scores for better indexing of the annotated data. By using a natural language processing-based term extraction measure, C-Value, we are able to enhance the original scoring algorithm which uses basic frequencies of the matches and in addition to positively discriminate multi-words term annotations. We show results obtained by comparing three different methods with a reference corpus of PubMed-MeSH manual annotations.

International audience; Approaches based on linguistic rules have been proposed to automatically extract candidate terms to help the terminology building from corpora. However, they face to the difficulty to identify the relevant terms among the noun phrases extracted. Although several statistical measures as the frequency or the C-Value have been proposed to ranked the terms according to their termhood, they fail to propose corpus and domain-independent ranking. We tackle this problem by proposing a parametrised C-Value which optimally considers the length and the syntactic roles of the nested terms thanks to a genetic algorithm. We compare its impact on the ranking of term extracted from on three corpora. Results show average precision increases by 9% above the frequency based ranking and by 12% above the C-Value based ranking.

International audience; In this paper, we study terminology extraction in all its states. We present two terminology extraction tools to compare a knowledge-poor vs. a knowledge-rich approach. Both tools deal with single and multi-word terms and are designed to handle multilingualism. Today they can process six languages, a new language can be added quite rapidly. We present an evaluation run on the six languages and two different domains using crawled comparable corpora, as well as hand-crafted reference term lists. We discuss the three main results for the terminology extraction issue. The first two scenarios concern the knowledge-rich framework. First we compare the performances of the ranking and compare the use of the specificity score vs. the the number of occurrences for all the languages. Secondly we examine the relevancy of the term variant identification to increase the precision ranking whatever the language is. The third evaluation scenario compares the two tools and shows that a probabilistic term extraction approach developed with a minimal effort achieves satisfactory results compared to a rule-based method.

Conference paper Extracting general or intermediate level terms is a relevant problem that has not received much attention in literature. Current approaches for term extraction rely on contrastive corpora to identify domain-specific terms, which makes them better suited for specialised terms, that are rarely used outside of the domain. In this work,we propose an alternative measure of domain specificity based on term coherence with an automatically constructed domain model. Although previous systems make use of domain-independent features, their performance varies across domains, while our approach displays a more stable behaviour, with results comparable to, or better than, state-of-the-art methods. EU Grant No. 258191 (PROMISE project); Science Foundation Ireland Grant No. SFI/12/RC/2289 peer-reviewed

International audience; This paper presents the work on terminology extraction from comparable corpora for Latvian. In the first section we introduce our work; the second section briefly describes the concept of the project and the implemented general terminology processing chain; the following two sections focus on terminology extraction workflow for Latvian and evaluation of results, respectively.

National audience; This paper describes the IRISA participation to the DeFT 2012 text-mining challenge. It consisted in the automatic attribution or generation of keywords to scientific journal articles. Two tasks were proposed which led us to test two different strategies. For the first task, a list of keywords was provided. Based on that, our first strategy is to consider that as an Information Retrieval problem in wich the keyword are the queries, which are attributed to the best ranked documents. This approach yielded very good results. For the second task, only the articles were known; for this task, our approach is chiefly based on a term extraction system whose results are reordered by machine learning.; Dans cet article, nous décrivons notre participation au Défi Fouille de Texte (DeFT) 2012. Ce défi consistait en l'attribution automatique de mots-clés à des articles scientifiques en français, selon deux pistes pour lesquelles nous avons employé des approches différentes. Pour la première piste, une liste de mots-clés était fournie. Nous avons donc abordé ce problème comme une tâche de recherche d'information dans laquelle les mots-clés sont les requêtes. Cette approche a donné d'excellents résultats. Pour la seconde piste, seuls les articles étant fournis, nous avons employé une approche s'appuyant sur un extracteur de terme et une réordonnancement par apprentissage.

International audience; In this paper, we exploit domain-specific comparable corpora to build bilingual terminologies. We present the monolingual term extraction and the bilingual alignment that will allow us to identify and translate high specialised terminology. We stress the huge importance of taking into account both simple and complex terms in a multilingual environment. Such linguistic diversity implies to combine several methods to perfect accurately both monolingual and bilingual terminology extraction tasks. The methods are implemented in TTC TermSuite based on a UIMA framework.

International audience; The Quæro program has organized a set of evaluations for terminology extraction systems in 2010 and 2011. Three objectives were targeted in this initiative: the first one was to evaluate the behavior and scalability of term extractors regarding the size of corpora, the second goal was to assess progress between different versions of the same systems, the last one was to measure the influence of corpus type. The protocol used during this initiative was a comparative analysis of 32 runs against a gold standard. Scores were computed using metrics that take into account gradual relevance. Systems produced by Quæro partners and publicly available systems were evaluated on pharmacology corpora composed of European Patents or abstracts of scientific articles, all in English. The gold standard was an unstructured version of the pharmacology thesaurus used by INIST-CNRS for indexing purposes. Most systems scaled with large corpora, contrasted differences were observed between different versions of the same systems and with better results on scientific articles than on patents. During the ongoing adjudication phase domain experts are enriching the thesaurus with terms found by several systems.

System Demonstrations; International audience; This paper aims at presenting TTC TermSuite: a tool suite for multilingual terminology extraction from comparable corpora. This tool suite offers a userfriendly graphical interface for designing UIMA-based tool chains whose components (i) form a functional architecture, (ii) manage 7 languages of 5 different families, (iii) support standardized file formats, (iv) extract single- and multi- word terms languages by languages (v) and align them by pairs of languages.

International audience; In this paper, we deal with bilingual terminology extraction from comparable corpora. The extraction can be seen as a pipeline of processing steps. We will discuss grouping of term variants and describe two methods for bilingual term alignment of neoclassical terms: a knowledge-poor approach using string similarity measures and a linguistically motivated approach which is extended to cover German compound nouns.

http://lodel.irevues.inist.fr/tralogy/index.php?id=91; International audience; This paper presents usage scenarios of the platform being developed within the TTC project (Terminology Extraction, Translation Tools and Comparable Corpora) along with the first feedback from potential users. The TTC project aims at leveraging translation tools, computer-assisted translation tools, and terminology management tools by automatically generating bilingual terminologies from comparable corpora in several languages of the European Union (English, French, German, Latvian and Spanish), as well as in Chinese and Russian. The TTC platform includes a web crawler and a corpora management tool, as well as tools for monolingual term extraction and bilingual terminology alignment, online terminology management, and terminology export into CAT tools and MT systems. Overall, the paper focuses on the language activities to be carried out with the TTC tools, issues with respect to the availability of required language resources and linguistic knowledge, and different user profiles and needs. Regarding potential user needs, we discuss the results of an online questionnaire-based survey on terminology and corpora issues conducted in the translation and localization industry to reveal user needs. Furthermore, we present the envisaged usage scenarios as well as first feedback from potential users. The expected TTC input and outputs are also outlined. Finally, as it seems clear that the amount of available data and resources will not be the same for all languages, we discuss technical solutions to achieve language coverage: the TTC tools will offer different approaches depending on the amount and type of linguistic knowledge available.

Poster and demonstration; National audience; In the TTC project, monolingual term candidate extraction tools are being developed, as a basis for later term alignment, using comparable corpora. TTC covers English, German, French, Spanish, Latvian, Russian and Chinese. In the poster, we address the extraction of morphological and syntactic term variants of the first four languages. The term extraction pipeline is aimed at texts of technical domains (e.g. wind energy), which are obtained by a thematic web-crawler and then are POS-tagged and lemmatized. We extract multiword term candidates by POS-patterns (e.g. adjective+noun) and statistically filter domain-relevant terms. In a second step, morpho-syntactic term variants are grouped into sets: (1) metallhaltiger Abfall ↔ Abfall aus Metall. We use the typology of term variation proposed by (Daille, 2005). Certain types of variation are captured by expanding basic POS-patterns; for modifications (adjectives, embedding under genitive or of-phrases), additional distributional data are used, e.g. the frequency of modifiers in particular constructions. For morphological variation, we experiment with a simple stemming-based approach and a morphological analyzer: stemming rules allow us to relate terms like production d'électricité and électricité produite. To relate e.g. Emissionssteigerung and gesteigerte Emission, we need to split the compound noun Emission|Steigerung and to model the relation between steigern and gesteigert, which can be done using SMOR (Schmid et al., 2004). In addition to monolingual term variants, we are interested in patterns applicable to several languages: (2) noun adjective = noun de noun énergie générable ↔ génération d'énergie (FR) energía generable ↔ generación de energía (ES).

International audience; The PERTOMed project is a pluri-disciplinary research initiative undertaken by several institutions in France. Applications considered within the part of the project described in this article concern pharmacovigilance and adverse drug reactions. We had multiple objectives: to create a specialized Russian Internet corpus; to test new tools and methods for term extraction from comparable multilingual texts and to build terminological resources including Russian. A trilingual Russian-French-English lexicon resulting form this work is freely available from the PERTOMed server.

The research work of this thesis is related to the problem of document search indexing and more specifically in that of the extraction of semantic descriptors for document indexing. Information Retrieval System (IRS) is a set of models and systems for selecting a set of documents satisfying user needs in terms of information expressed as a query. In IR, a query is composed mainly of two processes for representation and retrieval. The process of representation is called indexing, it allows to represent documents and query descriptors, or indexes. These descriptors reflect the contents of documents. The retrieval process consists on the comparison between documents representations and query representation. In the classical IRS, the descriptors used are words (simple or compound). These IRS consider the document as a set of words, often called a "bag of words". In these systems, the words are considered as graphs without semantics. The only information used for these words is their occurrence frequency in the documents. These systems do not take into account the semantic relationships between words. For example, it is impossible to find documents represented by a word synonymous with M1 word M2, where the request is represented by M2. Also, in a classic IRS document indexed by the term "bus" will never be found by a query indexed by the word "taxi", yet these are two words that deal with the same subject "means of transportation." To address these limitations, several studies were interested taking into account of the semantic indexing terms. This type of indexing is called semantic or conceptual indexing. These works take into account the notion of concept in place of notion of word. In this work the terms denoting concepts are extracted from the document by using statistical techniques. These terms are then projected onto resource of semantics such as: ontology, thesaurus and so on to extract the concepts involved.; Les travaux menés dans le cadre de cette thèse se situent dans la problématique de recherche- indexation des documents et plus spécifiquement dans celle de l’extraction des descripteurs sémantiques pour l’indexation. Le but de la Recherche d’Information (RI) est de mettre en œuvre un ensemble de modèles et de systèmes permettant la sélection d’un ensemble de documents satisfaisant un besoin utilisateur en termes d’information exprimé sous forme d’une requête. Un Système de Recherche d’Information (SRI) est composé principalement de deux processus. Un processus de représentation et un processus de recherche. Le processus de représentation est appelé indexation, il permet de représenter les documents et la requête par des descripteurs ou des indexes. Ces descripteurs reflètent au mieux le contenu des documents. Le processus de recherche consiste à comparer les représentations des documents à la représentation de la requête. Dans les SRIs classiques, les descripteurs utilisés sont des mots (simples ou composés). Ces SRIs considèrent le document comme étant un ensemble de mots, souvent appelé “ sac de mots ”. Dans ces systèmes, les mots sont considérés comme des graphies sans sémantique. Les seules informations exploitées concernant ces mots sont leurs fréquences d’apparition dans les documents. Ces systèmes ne prennent pas en considération les relations sémantiques entre les mots. Par exemple, il est impossible de trouver des documents représentés par un mot M1 synonyme d’un mot M2, dans le cas où la requête est représentée par M2. Aussi, dans un SRI classique un document indexé par le terme “ bus ” ne sera jamais retrouvé par une requête indexée par le terme “taxi ”, pourtant il s’agit de deux termes qui traitent le même thème “ moyen de transport ”. Afin de remédier à ces limites, plusieurs travaux se sont intéressés à la prise en compte de l’aspect sémantique des termes d’indexation. Ce type d’indexation est appelé indexation sémantique ou conceptuelle.

The research work of this thesis is related to the problem of document search indexing and more specifically in that of the extraction of semantic descriptors for document indexing. Information Retrieval System (IRS) is a set of models and systems for selecting a set of documents satisfying user needs in terms of information expressed as a query. In IR, a query is composed mainly of two processes for representation and retrieval. The process of representation is called indexing, it allows to represent documents and query descriptors, or indexes. These descriptors reflect the contents of documents. The retrieval process consists on the comparison between documents representations and query representation. In the classical IRS, the descriptors used are words (simple or compound). These IRS consider the document as a set of words, often called a "bag of words". In these systems, the words are considered as graphs without semantics. The only information used for these words is their occurrence frequency in the documents. These systems do not take into account the semantic relationships between words. For example, it is impossible to find documents represented by a word synonymous with M1 word M2, where the request is represented by M2. Also, in a classic IRS document indexed by the term "bus" will never be found by a query indexed by the word "taxi", yet these are two words that deal with the same subject "means of transportation." To address these limitations, several studies were interested taking into account of the semantic indexing terms. This type of indexing is called semantic or conceptual indexing. These works take into account the notion of concept in place of notion of word. In this work the terms denoting concepts are extracted from the document by using statistical techniques. These terms are then projected onto resource of semantics such as: ontology, thesaurus and so on to extract the concepts involved., Les travaux menés dans le cadre de cette thèse se situent dans la problématique de recherche- indexation des documents et plus spécifiquement dans celle de l’extraction des descripteurs sémantiques pour l’indexation. Le but de la Recherche d’Information (RI) est de mettre en œuvre un ensemble de modèles et de systèmes permettant la sélection d’un ensemble de documents satisfaisant un besoin utilisateur en termes d’information exprimé sous forme d’une requête. Un Système de Recherche d’Information (SRI) est composé principalement de deux processus. Un processus de représentation et un processus de recherche. Le processus de représentation est appelé indexation, il permet de représenter les documents et la requête par des descripteurs ou des indexes. Ces descripteurs reflètent au mieux le contenu des documents. Le processus de recherche consiste à comparer les représentations des documents à la représentation de la requête. Dans les SRIs classiques, les descripteurs utilisés sont des mots (simples ou composés). Ces SRIs considèrent le document comme étant un ensemble de mots, souvent appelé “ sac de mots ”. Dans ces systèmes, les mots sont considérés comme des graphies sans sémantique. Les seules informations exploitées concernant ces mots sont leurs fréquences d’apparition dans les documents. Ces systèmes ne prennent pas en considération les relations sémantiques entre les mots. Par exemple, il est impossible de trouver des documents représentés par un mot M1 synonyme d’un mot M2, dans le cas où la requête est représentée par M2. Aussi, dans un SRI classique un document indexé par le terme “ bus ” ne sera jamais retrouvé par une requête indexée par le terme “taxi ”, pourtant il s’agit de deux termes qui traitent le même thème “ moyen de transport ”. Afin de remédier à ces limites, plusieurs travaux se sont intéressés à la prise en compte de l’aspect sémantique des termes d’indexation. Ce type d’indexation est appelé indexation sémantique ou conceptuelle.

Back-of-the-book indexes are traditional devices that help readers to get access to document content. Such indexes present the book topics in a different form and order than in the document itself. In this paper, we show that back-of-the-book indexes, which give a synthetic view over the document content, are quite similar to indicative summaries. The new methods that are developed for the fine-grained indexing of documents can be compared with summarization technics. Besides their similarities, both types of tools belong to different traditions and the underlying methods are quite different. The confrontation helps to understand the specificity of each approach.

International audience; In this paper, we propose a pattern-based term extraction approach for Japanese, applying ACABIT system originally developed for French. The proposed approach evaluates termhood using morphological patterns of basic terms and term variants. After extracting term candidates, ACABIT system filters out non-terms from the candidates based on log-likelihood. This approach is suitable for Japanese term extraction because most of Japanese terms are compound nouns or simple phrasal patterns.

International audience; L’exploitation d’un grand ensemble de textes, dans le contexte d’une veille technologique par exemple, passe souvent par l’utilisation d’un logiciel de Text Mining. L’objet de cet article est de présenter un nouvel algorithme pour la construction de représentations simplifiées de contenus textuels. La simple notion de cooccurrence entre termes y est dépassée pour mettre en évidence des relations de nature hiérarchique, traduisant des différences de niveau de généralité au sein de la terminologie extraite d’un ensemble important de textes. Cet outil fait appel à un pré-traitement linguistique des textes permettant de dégager leur terminologie sous une forme normalisée, à la fois composée de termes simples et de multi-termes